Google has introduced a lower-priced option for its top generative artificial intelligence service and has said companies can save up to $1 billion annually by shifting a bulk of their work to AI.
The move comes as the US tech major aims to attract more users in a highly-competitive sector.
The Alphabet-owned company said its Gemini-powered Google AI Ultra subscription can now be accessed for $100 a month, it said at its annual Google I/O conference in Mountain View, California, on Tuesday.
That is significantly lower than the $250 it was charging for what was previously the only option for that service, but with a handful of features stripped. The full AI Ultra version, which comes with higher usage limits and substantial Google Cloud storage, is now down to $200.
Users and analysts had balked at the price of the initial service, especially given that its lower tiers, AI Plus and AI Pro, were offered at $8 and $20, respectively. All three tiers have been enhanced with additional features.
For comparison, the top service of Anthropic's Claude's starts at $100, while OpenAI's pioneering ChatGPT is at $200. DeepSeek, meanwhile, does not have a monthly subscription as it uses a pay-as-you-go model, and estimates show that heavy or professional use can cost as low as $25 a month.
It may also be a strategy to fend off the challenges posed by lower-priced generative AI services, most notably from China's DeepSeek, whose rock-bottom prices shook up the market, putting into question the true cost of developing large language models and even causing AI chip darling Nvidia's stock to crash by $600 billion last year.
Google, however, did not explicitly say this was the motivation for the price drop. Instead, the move taps into the momentum of Gemini, whose users have more than double to about 900 million in the past year, and the potential cost-efficiencies AI can offer, chief executive Sundar Pichai said.
“If companies used a mix [AI] models, they could save a lot of money. To put this in perspective, top companies are processing about one trillion tokens a day,” he told a media briefing ahead of Google I/O.
“If they shifted 80 per cent of their workloads, to a combination of 3.5 Flash and, and frontier models like 3.5 Pro … they'd save over $1 billion annually,” he said, referring to the Gemini model that boasts speed and “PhD-level reasoning”.
“You've probably heard anecdotes from other chief information officers that companies are already blowing through their annual token budgets … so it's real savings we're talking about that they can pour back into their company,” Mr Pichai added.
AI updates
At Google I/O, the California-based company unveiled Gemini 3.5 Flash, which it said is its most advanced offering for coding and agentic AI. It also introduced Gemini Omni, a follow-up to Nano Banana that is image and video-centric, capable of creating visuals with more accurate physics, Google said.
Google's bread-and-butter, Search, has also been updated with Gemini 3.5 Flash, with users now able to create multiple agents within the service.
On YouTube, the video-sharing platform was updated with Ask YouTube, which Google said will be able to generate results even from complex queries. The company also launched Docs Live, a voice-powered feature that helps structure a draft while pulling information from Google's services including Drive, Gmail and the web.
“In the future, you'll be able to create new docs and edit them directly all with your voice,” Mr Pichai said.
Google has significantly ramped up its investments in AI, as it aims to keep pace with other big AI companies, especially in an era where more companies and people increasingly adopt the technology.
The company had already said that it was increasing its AI capital expenditure to as high as $190 billion in 2026, markedly up from about $31 billion in 2022 – the same year ChatGPT was first made publicly available and the subsequent generative AI craze began to emerge.
A key part of that overall investment is Google's custom silicon chips. Mr Pichai said the company has taken a dual-chip approach with specialised architectures for training and other AI operations.
“We've taken a fundamentally different approach … [that] gives us the ability to create the largest training cluster in the world,” he said.
“This means training larger, more capable models in weeks rather than months.”
